Machine Learning seminars

The efficacy of images to create quantitative measures of urban perception has been explored in psychology, social science, urban planning and architecture over the last 50 years. The ability to scale these measurements has become possible only in the last decade, due to increased urban surveillance in the form of street view and satellite imagery, and the accessibility of such data. This talk will present a series of projects which make use of imagery and CNNs to predict, measure and interpret the social and physical environments of our cities.

As tiny robots become individually more sophisticated, and larger robots easier to mass produce, a breakdown of conventional disciplinary silos is enabling swarm engineering to be adopted across scales and applications, from nanomedicine to treat cancer, to cm-sized robots for large-scale environmental monitoring or intralogistics. This convergence of capabilities is facilitating the transfer of lessons learned from one scale to the other. Cm-sized robots that work in the 1000s may operate in a way similar to reaction-diffusion systems at the nanoscale, while sophisticated microrobots may have individual capabilities that allow them to achieve swarm behaviour reminiscent of larger robots with memory, computation, and communication. Although the physics of these systems are fundamentally different, much of their emergent swarm behaviours can be abstracted to their ability to move and react to their local environment. This presents an opportunity to build a unified framework for the engineering of swarms across scales that makes use of machine learning to automatically discover suitable agent designs and behaviours, digital twins to seamlessly move between the digital and physical world, and user studies to explore how to make swarms safe and trustworthy. Such a framework would push the envelope of swarm capabilities, towards making swarms for people.

In this workshop you'll learn the fundamentals of PyTorch using an incremental, from-first-principles approach. We'll start with tensors, autograd, and the dynamic computation graph, and then move on to developing and training a simple model using PyTorch's model classes, datasets, data loaders, optimizers, and more. You should be comfortable using Python, Jupyter notebooks, Google Colab, Numpy and, preferably, object oriented programming.

The success of deep learning sparked interest in whether the brain learns by using similar techniques for assigning credit to each synaptic weight for its contribution to the network output. However, the majority of current attempts at biologically-plausible learning methods are either non-local in time, require highly specific connectivity motives, or have no clear link to any known mathematical optimization method. Here, we introduce Deep Feedback Control (DFC), a new learning method that uses a feedback controller to drive a deep neural network to match a desired output target and whose control signal can be used for credit assignment. The resulting learning rule is fully local in space and time and approximates Gauss-Newton optimization for a wide range of feedback connectivity patterns. To further underline its biological plausibility, we relate DFC to a multi-compartment model of cortical pyramidal neurons with a local voltage-dependent synaptic plasticity rule, consistent with recent theories of dendritic processing. By combining dynamical system theory with mathematical optimization theory, we provide a strong theoretical foundation for DFC that we corroborate with detailed results on toy experiments and standard computer-vision benchmarks.

Networks are widely used as mathematical models of complex systems across many scientific disciplines, and in particular within neuroscience. In this talk, we introduce two aspects of our collaborative research: (1) machine learning and networks, and (2) graph dimensionality. Machine learning and networks. Decades of work have produced a vast corpus of research characterising the topological, combinatorial, statistical and spectral properties of graphs. Each graph property can be thought of as a feature that captures important (and sometimes overlapping) characteristics of a network. We have developed hcga, a framework for highly comparative analysis of graph data sets that computes several thousands of graph features from any given network. Taking inspiration from hctsa, hcga offers a suite of statistical learning and data analysis tools for automated identification and selection of important and interpretable features underpinning the characterisation of graph data sets. We show that hcga outperforms other methodologies (including deep learning) on supervised classification tasks on benchmark data sets whilst retaining the interpretability of network features, which we exemplify on a dataset of neuronal morphologies images. Graph dimensionality. Dimension is a fundamental property of objects and the space in which they are embedded. Yet ideal notions of dimension, as in Euclidean spaces, do not always translate to physical spaces, which can be constrained by boundaries and distorted by inhomogeneities, or to intrinsically discrete systems such as networks. Deviating from approaches based on fractals, here, we present a new framework to define intrinsic notions of dimension on networks, the relative, local and global dimension. We showcase our method on various physical systems.

A panel discussion on "What might we still require to achieve AGI?", a set of Reinforcement Learning and Computer Vision domain tuts and a talk from George Konidaris

Have a great idea involving AI? Want to launch your own business? It takes many iterations before an idea becomes a startup. Lots of coffee, heartache, and git reverts fuel these iterations. We have learned a lot from Cape AI's own incubated startup, Moonshop, Africa's first autonomous microstore. Watch the demo here: https://www.youtube.com/watch?v=odX6kxhLFC4 Attend our virtual roadshow event to hear lightning talks on creating proofs of concept, failing fast, funding models, selecting and growing a team, finding customers/clients, and building your brand. Afterwards, there will be a short break, then a panel discussion where members of the Cape AI team will answer questions from the audience.

Robust detection of biological structures such as neuronal dendrites in brightfield micrographs, tumor tissue in histological slides, or pathological brain regions in MRI scans is a fundamental task in bio-image analysis. Detection of those structures requests complex decision making which is often impossible with current image analysis software, and therefore typically executed by humans in a tedious and time-consuming manual procedure. Supervised pixel classification based on Deep Convolutional Neural Networks (DNNs) is currently emerging as the most promising technique to solve such complex region detection tasks. Here, a self-learning artificial neural network is trained with a small set of manually annotated images to eventually identify the trained structures from large image data sets in a fully automated way. While supervised pixel classification based on faster machine learning algorithms like Random Forests are nowadays part of the standard toolbox of bio-image analysts (e.g. Ilastik), the currently emerging tools based on deep learning are still rarely used. There is also not much experience in the community how much training data has to be collected, to obtain a reasonable prediction result with deep learning based approaches. Our software YAPiC (Yet Another Pixel Classifier) provides an easy-to-use Python- and command line interface and is purely designed for intuitive pixel classification of multidimensional images with DNNs. With the aim to integrate well in the current open source ecosystem, YAPiC utilizes the Ilastik user interface in combination with a high performance GPU server for model training and prediction. Numerous research groups at our institute have already successfully applied YAPiC for a variety of tasks. From our experience, a surprisingly low amount of sparse label data is needed to train a sufficiently working classifier for typical bioimaging applications. Not least because of this, YAPiC has become the "standard weapon” for our core facility to detect objects in hard-to-segement images. We would like to present some use cases like cell classification in high content screening, tissue detection in histological slides, quantification of neural outgrowth in phase contrast time series, or actin filament detection in transmission electron microscopy.

Several excellent computational frameworks exist that enable high-throughput and consistent tracking of freely moving unmarked animals. SimBA introduce and distribute a plug-and play pipeline that enables users to use these pose-estimation approaches in combination with behavioral annotation for the generation of supervised machine-learning behavioral predictive classifiers. SimBA was developed for the analysis of complex social behaviors, but includes the flexibility for users to generate predictive classifiers across other behavioral modalities with minimal effort and no specialized computational background. SimBA has a variety of extended functions for large scale batch video pre-processing, generating descriptive statistics from movement features, and interactive modules for user-defined regions of interest and visualizing classification probabilities and movement patterns.

There has been a recent surge of interest in the question of whether and how deep learning algorithms might be capable of abstract reasoning, much of which has centered around datasets based on Raven’s Progressive Matrices (RPM), a visual analogy problem set commonly employed to assess fluid intelligence. This has led to the development of algorithms that are capable of solving RPM-like problems directly from pixel-level inputs. However, these algorithms require extensive direct training on analogy problems, and typically generalize poorly to novel problem types. This is in stark contrast to human reasoners, who are capable of solving RPM and other analogy problems zero-shot — that is, with no direct training on those problems. Indeed, it’s this capacity for zero-shot reasoning about novel problem types, i.e. fluid intelligence, that RPM was originally designed to measure. I will present some results from our recent efforts to model this capacity for zero-shot reasoning, based on an extension of a recently proposed approach to analogical mapping we refer to as Probabilistic Analogical Mapping (PAM). Our RPM model uses deep learning to extract attributed graph representations from pixel-level inputs, and then performs alignment of objects between source and target analogs using gradient descent to optimize a graph-matching objective. This extended version of PAM features a number of new capabilities that underscore the flexibility of the overall approach, including 1) the capacity to discover solutions that emphasize either object similarity or relation similarity, based on the demands of a given problem, 2) the ability to extract a schema representing the overall abstract pattern that characterizes a problem, and 3) the ability to directly infer the answer to a problem, rather than relying on a set of possible answer choices. This work suggests that PAM is a promising framework for modeling human zero-shot reasoning.

Hongjing Lu will present a new computational model of Probabilistic Analogical Mapping (PAM, in collaboration with Nick Ichien and Keith Holyoak) that finds systematic correspondences between inputs generated by machine learning. The model adopts a Bayesian framework for probabilistic graph matching, operating on semantic relation networks constructed from distributed representations of individual concepts (word embeddings created by Word2vec) and of relations between concepts (created by our BART model). We have used PAM to simulate a broad range of phenomena involving analogical mapping by both adults and children. Our approach demonstrates that human-like analogical mapping can emerge from comparison mechanisms applied to rich semantic representations of individual concepts and relations. More details can be found https://arxiv.org/ftp/arxiv/papers/2103/2103.16704.pdf

Animals produce behavior by responding to a mixture of cues that arise both externally (sensory) and internally (neural dynamics and states). These cues are continuously produced and can be combined in different ways depending on the needs of the animal. However, the integration of these external and internal cues remains difficult to understand in natural behaviors. To address this gap, we have developed an unsupervised method to identify internal states from behavioral data, and have applied it to the study of a dynamic social interaction. During courtship, Drosophila melanogaster males pattern their songs using cues from their partner. This sensory-driven behavior dynamically modulates courtship directed at their partner. We use our unsupervised method to identify how the animal integrates sensory information into distinct underlying states. We then use this to identify the role of courtship neurons in either integrating incoming information or directing the production of the song, roles that were previously hidden. Our results reveal how animals compose behavior from previously unidentified internal states, a necessary step for quantitative descriptions of animal behavior that link environmental cues, internal needs, neuronal activity, and motor outputs.

(1) We (Grace Lindsay) used convolutional neural nets to model attention, by scaling the input/output function of neurons in an imagenet-trained network according to their selectivity for the feature or object category being attended. While this was effective in improving performance on difficult tasks, it was far less effective in earlier than in later layers. This indicated that neurons selective for a feature in earlier layers did not necessarily drive neurons selective for that feature in later layers. In contrast, applying attention according to the gradient for improving task performance worked well in early as well as late layers. This raises the question whether biological attentional modulation might reflect task requirements and not only the features of the stimuli to be attended. We suggest a simple experiment to answer this question, which we hope to convince an appropriate lab to carry out. (2) In E/I networks, a "paradoxical" response to stimulation has been shown: If the excitatory neurons would be unstable by themselves, but are stabilized by feedback inhibition (an "inhibition-stabilized network", or ISN), then, in response to addition of excitatory input to inhibitory neurons, their steady-state firing rates paradoxically decrease. In circuits with multiple inhibitory cell types, this has been generalized: in an ISN, if there is an added stimulus only to inhibitory cells, there will be a paradoxical change in the net inhibition received by excitatory cells -- e.g., if excitatory firing rates increase, so too will the net inhibition they receive. This does not imply that the firing rates of any particular inhibitory cell type will change paradoxically. Here we (Agostina Palmigiano along with Francesco Fumarola, and experimental work of Dan Mossing in the Adesnik lab) generalize the conditions for a paradoxical firing rate response, including in responses to partial as well as full perturbation of the neurons of a given cell type. We work in the context of the circuit with three inhibitory cell types (PV, SOM, VIP) in mouse V1. We show that, if a given cell type shows a paradoxical response to its own full stimulation, then the circuit without that cell type is unstable. This and experimental results to date, as well as our models fitted to data, suggest that PV but not SOM interneurons stabilize the circuit of layer 2/3 of mouse V1, at least for smaller visual stimulus sizes. For partial perturbations of a fraction f of a cell type that responds paradoxically to a full perturbation, there is a "fractional paradoxical effect": the proportion of all the cells of that type, stimulated and unstimulated, that respond opposite to the stimulation (i.e. negative response to excitation), changes non-monotonically, approaching 1 for f→0, decreasing with increasing f, and then increasing again to again approach 1 as f→1. I'll explain the origins of this behavior.3) We (Mario Dipoppa, in collaboration with the experimental work of Andy Keller and Morgane Roth from the Scanziani lab) have studied the E-PV-SOM-VIP circuit underlying contextual modulation in layer 2/3 of mouse V1. Experiments showed that E, PV, and VIP are suppressed by a surround stimuus that has the same orientation as, but not by one orthogonal to, the center stimulus. SOM neurons show the opposite behavior, being suppressed by an orthogonal but much less by a parallel surround. A combination of theory and optogenetic experiments show that the disinhibitory circuit -- VIP inhibits SOM, which inhibits E -- modulate responses between the two conditions. However, it does so, as part of the recurrent circuit, primarily by changing the recurrent excitation E cells receive, rather than by directly changing the inhibition received, in a manner reminiscent of the paradoxical response. Presented in the van Vreeswijk Theoretical Neuroscience Seminar series (formerly WWTNS) on 2021-06-09. Recording duration: 00:51:59.

Despite the importance of early diagnosis of dementia for prognosis and personalised interventions, we still lack robust tools for predicting individual progression to dementia. We propose a trajectory modelling approach that mines multimodal data from patients at early dementia stages to derive individualised prognostic scores of cognitive decline Our approach has potential to facilitate effective stratification of individuals based on prognostic disease trajectories, reducing patient misclassification with important implications for clinical practice.

The Stabilized Supralinear Network is a model of recurrently connected excitatory (E) and inhibitory (I) neurons with a supralinear input-output relation. It can explain cortical computations such as response normalization and inhibitory stabilization. However, the network's connectivity is designed by hand, based on experimental measurements. How the recurrent synaptic weights can be learned from the sensory input statistics in a biologically plausible way is unknown. Earlier theoretical work on plasticity focused on single neurons and the balance of excitation and inhibition but did not consider the simultaneous plasticity of recurrent synapses and the formation of receptive fields. Here we present a recurrent E-I network model where all synaptic connections are simultaneously plastic, and E neurons self-stabilize by recruiting co-tuned inhibition. Motivated by experimental results, we employ a local Hebbian plasticity rule with multiplicative normalization for E and I synapses. We develop a theoretical framework that explains how plasticity enables inhibition balanced excitatory receptive fields that match experimental results. We show analytically that sufficiently strong inhibition allows neurons' receptive fields to decorrelate and distribute themselves across the stimulus space. For strong recurrent excitation, the network becomes stabilized by inhibition, which prevents unconstrained self-excitation. In this regime, external inputs integrate sublinearly. As in the Stabilized Supralinear Network, this results in response normalization and winner-takes-all dynamics: when two competing stimuli are presented, the network response is dominated by the stronger stimulus while the weaker stimulus is suppressed. In summary, we present a biologically plausible theoretical framework to model plasticity in fully plastic recurrent E-I networks. While the connectivity is derived from the sensory input statistics, the circuit performs meaningful computations. Our work provides a mathematical framework of plasticity in recurrent networks, which has previously only been studied numerically and can serve as the basis for a new generation of brain-inspired unsupervised machine learning algorithms.

In vivo, the human genome folds into a characteristic ensemble of 3D structures. The mechanism driving the folding process remains unknown. A theoretical model for chromatin (the minimal chromatin model) explains the folding of interphase chromosomes and generates chromosome conformations consistent with experimental data is presented. The energy landscape of the model was derived by using the maximum entropy principle and relies on two experimentally derived inputs: a classification of loci into chromatin types and a catalog of the positions of chromatin loops. This model was generalized by utilizing a neural network to infer these chromatin types using epigenetic marks present at a locus, as assayed by ChIP-Seq. The ensemble of structures resulting from these simulations completely agree with HI-C data and exhibits unknotted chromosomes, phase separation of chromatin types, and a tendency for open chromatin to lie at the periphery of chromosome territories. Although this theoretical methodology was trained in one cell line, the human GM12878 lymphoblastoid cells, it has successfully predicted the structural ensembles of multiple human cell lines. Finally, going beyond Hi-C, our predicted structures are also consistent with microscopy measurements. Analysis of both structures from simulation and microscopy reveals that short segments of chromatin make two-state transitions between closed conformations and open dumbbell conformations. For gene active segments, the vast majority of genes appear clustered in the linker region of the chromatin segment, allowing us to speculate possible mechanisms by which chromatin structure and dynamics may be involved in controlling gene expression. * Supported by the NSF

In the Learning Salon, we will discuss the similarities and differences between biological and machine learning, including individuals with diverse perspectives and backgrounds, so we can all learn from one another.

Humans can adapt to a novel task on our first try. By contrast, artificial intelligence systems often require immense amounts of data to adapt. In this talk, I will discuss my recent work (https://www.pnas.org/content/117/52/32970) on creating deep learning systems that can adapt on their first try by exploiting relationships between tasks. Specifically, the approach is based on transforming a representation for a known task to produce a representation for the novel task, by inferring and then using a higher order function that captures a relationship between the tasks. This approach can be interpreted as a type of analogical reasoning. I will show that task transformation can allow systems to adapt to novel tasks on their first try in domains ranging from card games, to mathematical objects, to image classification and reinforcement learning. I will discuss the analogical interpretation of this approach, an analogy between levels of abstraction within the model architecture that I refer to as homoiconicity, and what this work might suggest about using deep-learning models to infer analogies more generally.

In the Learning Salon, we will discuss the similarities and differences between biological and machine learning, including individuals with diverse perspectives and backgrounds, so we can all learn from one another.

DeepLabStream is a python based multi-purpose tool that enables the realtime tracking and manipulation of animals during ongoing experiments. Our toolbox was orginally adapted from the previously published DeepLabCut (Mathis et al., 2018) and expanded on its core capabilities, but is now able to utilize a variety of different network architectures for online pose estimation (SLEAP, DLC-Live, DeepPosekit's StackedDenseNet, StackedHourGlass and LEAP). Our aim is to provide an open-source tool that allows researchers to design custom experiments based on real-time behavior-dependent feedback. My personal ideal goal would be a swiss-army knife like solution where we could integrate the many brilliant python interfaces. We are constantly upgrading DLStream with new features and integrate other open-source solutions.